Data analysis - catchment nutrient reports

All data used in these nutrient reports are available for download at:

Prior to analysis, all values that were less than the limit of reporting were assigned a value that was half the limit of reporting.

1.0 Comparison to targets and trigger values

Water quality variables analysed were compared with a range of targets and trigger values. These were used as a point of reference against which to compare data points both within and between sites. Water quality was also classified using a rigorous statistical technique which reduces the effects of inter-annual variability and allows us to track changes in water quality over time (see section 2.0). This was used to assign a status for each site and variable. The most appropriate targets and trigger values for each estuary catchment were selected for comparison. These are outlined in the following sections.

2.0 Variable classification (modified from Hall 2010)

The total nitrogen, total phosphorus, dissolved organic carbon, total suspended solids and salinity concentrations at all catchment sites sampled as part of the Health Estuaries WA sampling programs were classified using the classification bands shown below. All classification bands (with the exception of salinity) are from the Statewide River Water Quality Assessment webpage (SWRWQA 2009). To classify the salinity data, the Water Resources Inventory 2014 salinity ranges (Department of Water 2014) were used.

Table 1: classification bands for total nitrogen, total phosphorus, dissolved organic carbon and total suspended solids (from SWRWQA 2009). If viewing on a mobile device, view table here.


Total nitrogen (mg/L)

Total phosphorus (mg/L)

Dissolved organic carbon (mg/L)

Total suspended solids (mg/L)

very high

> 2

> 0.20

> 25

> 25


> 1.2 – 2

> 0.08 – 0.20

> 10 – 25

> 10 – 25


0.75 – 1.2

0.02 – 0.08

5 – 10

5 – 10


< 0.75

< 0.02

< 5

< 5

Table 2: classification bands for salinity (from Department of Water 2014). If viewing on a mobile device, view table here.


Salinity (mg/L)


> 3,000


> 1,000 – 3,000


500 – 1,000


< 500

Depending on trends, chance sampling and sources of natural variation, the nutrient concentrations analysed from a monitored site will change. The nutrient status for a waterway is initially assigned using the median nutrient concentration for the first year of sampling. Subsequent status periods are assessed using the median and 90% confidence interval. If the median or all or part of the confidence interval remains in the earlier classification band, then there is no change in status. Status only changes once both the median and entire 90% confidence interval move to a different classification band.

Figure 1 shows how TN status at Mayfields Main Drain (in the Peel-Harvey catchment) was originally classified as high (the median was between 1.2 and 2.0 mg/L). By the 1992–94 period, the median had decreased and fell within the moderate classification band (0.75–1.2 mg/L); however, part of the 90% confidence interval was still in the high classification band and so the status remained high. In the 1994–96 period, both the median and 90% confidence interval fell below the high classification and hence the status changed to moderate. During the 1996–98 period the median once again dropped to a lower classification band (<0.75 mg/L); however, it wasn’t until the 1998–2000 period that the actual classification status changed to low.

In summary, the nutrient status for a waterway is assigned by using the median of nutrient concentration over a three-year period. The three-year period is used to reduce the influence of natural variation between years. Change in status requires the median and whole 90% confidence interval to be within the new status concentration range.

Figure 1: Total phosphorus status classification for Mayfields Main Drain (AWRC 613031)

Figure 1: Total nitrogen status classification for Mayfields Main Drain (AWRC 613031)

3.0 Statistical trend testing methodology (modified from Hall 2010)

3.1 Testing for statistically significant changes

The Mann-Kendall test is used to determine the statistical significance of the trends in water quality over time (Gilbert 1987). It is a non-parametric test and is only used when the data series exhibits independence (i.e. no correlation in the data series) (Figure 2). The Mann-Kendall test works by calculating a statistic ‘S’ and testing the significance of this statistic. Each data pair is compared and assigned a plus or a minus depending on whether the later data point is higher than the earlier data point. ‘S’ is the overall number of pluses or minuses (where one plus cancels out one minus) for the whole dataset (Nelson 2004). The Z-statistic, from which the ‘p-value’ is derived, is calculated as follows:

Where Var(S) is the variance of the dataset used to derive ‘S’. An increasing trend will have a large positive Z-statistic, while the Z statistic for a decreasing trend will be negative and have a large absolute value.


Figure 2: Example of a time-series with little evidence of a seasonal pattern in total phosphorus concentration, hence the Mann-Kendall test for trend is used

Seasonal cycles in nutrient concentration are common in waterways and can be introduced by natural cycles in rainfall, runoff, tributary hydrology and seasonal variation in groundwater interaction. When seasonal cycles are evident in a data series (Figure 3), the Seasonal-Kendall test is used to test for trend. The Seasonal-Kendall test is a variant of the Mann-Kendall test that accounts for the presence of seasonal cycles in the data series (Gilbert 1987). The ‘S’ statistic is calculated slightly differently in the Seasonal-Kendall test. Rather than comparing all data pairs, only data points falling in the same ‘season’ are compared. For example, if a weekly season is used, data points from the first weeks of the year are only compared with data points from the first week of all other years.


Figure 3: An example of a pronounced seasonal pattern in total phosphorus concentration

Nutrient concentrations in waterways can also be affected by changes in flow. The relationship between nutrient concentration and flow is modelled using LOWESS fit between the concentration and flow (Helsel & Hirsch 1992). The difference of ‘residuals’ between the observed and LOWESS modelled concentration are termed flow-adjusted concentrations (FAC), as shown in Figure 5-9 (Hipel & McLeod 1994). Trend analyses may then be performed on the flow-adjusted concentrations. The flow-adjustment process often helps to remove seasonal variation (as shown by comparing Figures 3 and 4B), although some evidence of seasonal variation often remains in the flow-adjusted data series.


Figure 4: The flow response plot shows whether a relationship exists between discharge and nutrient concentration (A). The flow-adjusted concentrations (or residuals) are the difference between observed and modelled (LOWESS) concentrations (B).

3.2 Estimating the rate of change

The Sen slope estimator is used to estimate the slope of the trend line (Gilbert 1987). The Sen estimate is calculated in a similar manner to the test statistic ‘S’ from the Mann-Kendall test. Rather than comparing each data pair from an increase or decrease over time, a slope is calculated using each data pair. The Sen slope estimator is taken as the median slope of all slopes calculated using all data pairs. In the presence of seasonal cycles the Seasonal-Kendall slope estimator is used. This is similar to the seasonal test ‘S’ in the Seasonal-Kendall test, in that slopes are only calculated for data pairs from the same season. The Sen slope estimator is the median of all these slopes. Figure 5 shows an example of a slope estimated for a series showing seasonally.


Figure 5: An example of how the Seasonal Sen slope estimator represents the slope of the trend line in a seasonal nutrient data series.

4.0 Estimating loads

Loads were calculated for those sites where there was sufficient flow and nutrient data available.

Annual loads were calculated by multiplying daily flow with daily nutrient concentration and aggregating over the year. Daily concentration measurements are not available as samples were taken weekly at most, so daily concentration data needs to be in-filled to calculate loads. To calculate the in-filled nutrient data the locally estimated scatterplot smoothing (LOESS) algorithm (Cleveland 1979) was used.

LOESS creates a flow-concentration curve by fitting a low-degree polynomial to a subset of the flow-concentration data to estimate the concentration for the flow at the centre point of the data subset. This is done for each flow value in the dataset. For days on which nutrient data were collected, daily loads are calculated from observed concentrations and flows. For days with no data, daily loads are calculated from the daily flow and the estimated concentration from the LOESS flow-concentration curve. The assumption of the LOESS algorithm is that there is a relationship between flow and concentration.


ANZECC & ARMCANZ, 2000, Australian and New Zealand guidelines for fresh and marine water quality, Australian and New Zealand Environment and Conservation Council & Agriculture and Resource Management Council of Australia and New Zealand.

Cleveland, WS 1979, Robust locally weighted regression and smoothing scatterplots, Journal of the American Statistical Association, Vol 74: 829-836.

Department of Water, 2014, WAter resources inventory 2014. Water availability, quality and trends, Department of Water, Western Australia.

Department of Water and Environmental Regulation, 2022, Bindjareb Djilba, a plan for the protection of the Peel-Harvey estuary, Department of Water and Environmental Regulation, Western Australia.

Gilbert, R 1987, Statistical methods for environmental pollution monitoring, Van Nostrand Reinhold, New York, 250 pp.

Hall, J 2010, Water quality management in urban catchments of the Swan Coastal Plain: analysis of the Bartram Road catchment, Water Science Technical Series, report no. 22, Department of Water, Western Australia.

Hipel, K & McLeod, A 1994, Time series modelling of environmental and water resources systems, Elsevier, Amsterdam.

Hugues-dit-Ciles, J, Kelsey, P, Marillier, B, Robb, M, Forbes, V & McKenna, M 2012, Leschenault estuary water quality improvement plan, Department of Water, Western Australia.

Nelson, S 2004, Planet (Version 5) user manual, Water and Rivers Commission, Perth, Western Australia.

SWRWQA 2009, Statewide river water quality assessment; accessed: 31/03/2022.

White, KS 2012, Hardy Inlet water quality improvement plan: Stage one – the Scott River catchment; Department of Water, Western Australia.